[DPE-4739] Avoid using time.sleep in rollback integration tests (#298)

## Issue
1. We use time.sleep instead of a deterministic way to determine when to
send `resume-upgrade` after rolling back. This sometimes leads to [this
error](https://github.com/canonical/mysql-router-k8s-operator/actions/runs/10312040824/job/28546894902#step:31:2522)
when `resume-upgrade` is run too early
```
Resume upgrade event failed: Highest number unit is unhealthy. Upgrade will not resume.
```
2. We are using `prime` in charmcraft which is deprecated in charmcraft
3.1.1

## Solution
1. Use a more deterministic approach - check that all units are `active`
and `idle` (one will have rolled back while the other two will not)
2. Use parts to prime version files and the dump plugin to prime
scripts/ directory

charmcraft.yaml

@@ -10,6 +10,13 @@ bases:
     channel: "22.04"
     architectures: [arm64]
 parts:
+  files:
+    plugin: dump
+    source: .
+    prime:
+      - charm_version
+      - workload_version
+      - scripts
   charm:
     override-pull: |
       craftctl default
@@ -20,10 +27,6 @@ parts:
       fi
     charm-strict-dependencies: true
     charm-entrypoint: src/kubernetes_charm.py
-    prime:
-      - charm_version
-      - workload_version
-      - scripts
     build-packages:
       - libffi-dev
       - libssl-dev